Benchmark was refactored to not overflow memory

[svetlyak40wt]

Now threads generating load are pause generation if actor's queue size growth above given threshold.

Now benchmark uses trivial-benchmark library which formats results in a unified way on different lisp implementations. However the most wide variety of metrics are supported for SBCL only.

The pro of using trivial-benchmark here is that it automates repeated tests. Now you can set a number of runs for the benchmark and a duration of the each run. And trivial-benchmark will automatically calculate average, median and deviation of each metric across all runs.

For Sento benchmark I've added a special metric - a message per second. During my tests satisfactory results were obtained when test duration was set to 10 seconds and a number of runs to 60.

On my Macbook M1 with 16G of memory benchmark results are:

SENTO/BENCH> (lisp-implementation-type)
"SBCL"
SENTO/BENCH> (lisp-implementation-version)
"2.5.0.116-bd18806aa"
SENTO/BENCH> (uiop:get-optimization-settings)
((SPEED 1) (SPACE 1) (SAFETY 1) (DEBUG 1) (COMPILATION-SPEED 1))


Results for benchmark: (:DISPATCHER :PINNED :WITH-REPLY-P NIL :ASYNC-ASK-P NIL
                        :NUM-SHARED-WORKERS 8 :QUEUE-SIZE NIL
                        :WAIT-IF-QUEUE-LARGE-THAN 10000)
┌─────────────────────┬───────────────┬─────────────┬─────────────┬─────────────┬─────────────┬─────────────┐
│ -                   │ TOTAL         │ MINIMUM     │ MAXIMUM     │ MEDIAN      │ AVERAGE     │ DEVIATION   │
├─────────────────────┼───────────────┼─────────────┼─────────────┼─────────────┼─────────────┼─────────────┤
│ MESSAGES-PER-SECOND │    59763090.0 │    657702.9 │   1103853.4 │   1014389.2 │   996051.44 │    76637.07 │
├─────────────────────┼───────────────┼─────────────┼─────────────┼─────────────┼─────────────┼─────────────┤
│ USER-RUN-TIME       │      707.3372 │    7.839694 │   12.719797 │   11.985142 │   11.788954 │    0.839375 │
├─────────────────────┼───────────────┼─────────────┼─────────────┼─────────────┼─────────────┼─────────────┤
│ SYSTEM-RUN-TIME     │     232.37593 │    2.988574 │    4.180502 │    3.896854 │    3.872932 │    0.177386 │
├─────────────────────┼───────────────┼─────────────┼─────────────┼─────────────┼─────────────┼─────────────┤
│ REAL-TIME           │      635.9478 │      10.508 │      10.637 │      10.608 │   10.599131 │    0.029722 │
├─────────────────────┼───────────────┼─────────────┼─────────────┼─────────────┼─────────────┼─────────────┤
│ GC-RUN-TIME         │         3.924 │        0.05 │       0.075 │       0.067 │      0.0654 │    0.005194 │
├─────────────────────┼───────────────┼─────────────┼─────────────┼─────────────┼─────────────┼─────────────┤
│ GC-REAL-TIME        │         4.052 │        0.05 │       0.128 │       0.067 │    0.067533 │    0.009718 │
├─────────────────────┼───────────────┼─────────────┼─────────────┼─────────────┼─────────────┼─────────────┤
│ PROCESSOR-CYCLES    │           0.0 │         0.0 │         0.0 │         0.0 │         0.0 │         0.0 │
├─────────────────────┼───────────────┼─────────────┼─────────────┼─────────────┼─────────────┼─────────────┤
│ EVAL-CALLS          │       26520.0 │       442.0 │       442.0 │       442.0 │       442.0 │         0.0 │
├─────────────────────┼───────────────┼─────────────┼─────────────┼─────────────┼─────────────┼─────────────┤
│ LAMBDAS-CONVERTED   │           0.0 │         0.0 │         0.0 │         0.0 │         0.0 │         0.0 │
├─────────────────────┼───────────────┼─────────────┼─────────────┼─────────────┼─────────────┼─────────────┤
│ PAGE-FAULTS         │           0.0 │         0.0 │         0.0 │         0.0 │         0.0 │         0.0 │
├─────────────────────┼───────────────┼─────────────┼─────────────┼─────────────┼─────────────┼─────────────┤
│ BYTES-CONSED        │ 121484440000. │ 1338653400. │ 2245857000. │ 2063355600. │ 2024740700. │ 155061730.0 │
└─────────────────────┴───────────────┴─────────────┴─────────────┴─────────────┴─────────────┴─────────────┘

Note, you will need the latest trivial-benchmark with this PR merged. It is available from https://ultralisp.org

[mdbergmann]

Looks fine.

[mdbergmann]

I've made a few minor adjustments.
"serapeum" was missing as dependency of bench system.
I've added a dedicated variable for tracking queue fill count and this is queried instead of locking and calculating plus traversing the head and tail lists.